Abstract
When the task is reading nonwords aloud, skilled adult readers are very variable in the responses they produce: a nonword can evoke as many as 24 different responses in a group of such readers. Why is nonword reading so variable? We analysed a large database of reading responses to nonwords, which documented that two factors contribute to this variability. The first factor is variability in graphemic parsing (the parsing of a letter string into its constituent graphemes): the same nonword can be graphemically parsed in different ways by different readers. The second factor is phoneme assignment: even when all subjects produce the same graphemic parsing of a nonword, they vary in what phonemes they assign to the resulting set of graphemes. We consider the implications of these results for the computational modelling of reading, for the assessment of impairments of nonword reading, and for the study of reading aloud in other alphabetically written languages and in nonalphabetic writing systems.